Skip to main content

DoS protection

Available since @subsquid/graphql-server@2.1.0

The squid GraphQL API server accepts the following optional start arguments to fend off heavy queries.

To enable the protection for squids deployed to Aquarium, add the corresponding flags the GraphQL api service command in the deployment manifest. Here is an example:

squid.yaml
# ...
deploy:
# other services ...
api:
cmd: [ "npx", "squid-graphql-server", "--max-root-fields", "10", "--max-response-size", "1000" ]

For local development, update commands.json accordingly:

...
"serve": {
"description": "Start the GraphQL API server",
"cmd": ["squid-graphql-server", "--max-root-fields", "10", "--max-response-size", "1000"]
},
...

--max-request-size <kb>

The argument limits the size of a request in kilobytes. It is set to 256kb by default.

--max-root-fields <count>

The maximal allowed number of root-level queries in a single GraphQL request.

--max-response-size <nodes>

This option limits the estimated query response size and makes server return an error if it exceeds the provided value. Note that the estimated size depends only on the decorators in schema.graphql and the requested fields.

The estimate is the product of the cardinality of the entity list and the response item weight.

The cardinality is estimated as the minimum of

  • the limit argument of the query (Infinity if not provided)
  • @cardinality value defined in schema.graphql (if the requested entity type is decorated in the schema file, Infinity otherwise)
  • the size of the argument list of the _eq and id_in filters in the where clause (if applicable)

In particular, if there are no @cardinality decorators in schema.graphql, the client queries must explicitly provide limits or where filters to pass through.

The response item weight is calculated recursively:

  • byteWeight for each scalar field or 1 if it's not decorated
  • for non-scalar fields, the estimated weight times the estimated cardinality (if it's a list)
  • each non-leaf node in the query AST tree adds a weight of 1

In a nutshell, assuming that the schema file is properly decorated with @cardinality and @byteWeight, the estimated response size should roughly be at the same scale as the byte size of the query result.

--subscription-max-response-size <nodes>

Same as --max-response-size but for live query subscriptions.

Example‚Äč

Assume the schema is defined as follows, and the server is launched with --max-response-size 1000.

schema.graphql
type Foo @entity {
id: ID!
// a large string
bigField: String! @byteWeight(value: 1000.0)
bar: Bar!
}

type Bar @entity {
id: ID!
// bar.foos typically contain about 100 items
foos: [Foo!]! @derivedFrom(field: "bar") @cardinality(value: 100)
bazs: [Baz!]! @derivedFrom(field: "bar")
}

// there are around 100 entities of type Baz
type Baz @entity @cardinality(value: 100) {
id: ID!
bar: Bar!
}

The following queries will be bounced:

query A {
bars {
id
}
}

query B {
bars(limit: 1001) {
id
}
}

query C {
bars(limit: 100) {
id
foos(limit: 10) {
id
}
}
}

query D {
bars(limit: 10) {
id
foos {
id
}
}
}

query E {
foos(id_eq: "1") {
id
bigField
}
}
  • The estimated cardinality of query A is Infinity
  • The estimated cardinality of query B 1001 and so the expected size exceeds the limit
  • The estimated cardinality of query C is 100 while the item size is 13, so the size is estimated to 1300.
  • The estimated cardinality of query D is 10 while the item size is 103, so the size is estimated to 1030.
  • The estimated cardinality of query E is 1 while the item size is 1001 (due to bigField having weight 1000).

At the same time, the following queries will go through:

query A {
bars(limit: 100) {
id
}
}

query B {
bars(limit: 3) {
id
foos {
id
}
bazs {
id
}
}
}

query C {
bars {
id
bar {
foos(where: { id_in ["1", "2" ]}) {
id
}
id
}
}
}